RekomBeli : Product Recommendation on e-Commerce “KlikBeli”

12 min readJul 9, 2023

Case Description

2023 is an exciting year for the company “KlikBeli”, an in-demand e-commerce platform. With great passion, “KlikBeli” is committed to providing customers with quality products. Over time, “KlikBeli” realized that there were significant challenges in sharing effective product promotions to increase the number of purchases. “KlikBeli” faced several challenges, namely the lack of a deep understanding of customer preferences and purchasing behavior. This makes it difficult for the company to determine the most attractive offers for each customer individually. In addition, “KlikBeli” struggles to manage and analyze large and complex customer data so the company does not have an efficient way to deliver promotions to customers. As a result, customers do not respond well to the promotions that have been carried out. Many customers are not interested or feel that the promotion is not relevant to them.

The Problem

The low sales conversion is due to the small number of customers who make purchases after receiving promotions. “KlikBeli” lacks the ability to analyze large customer data so that promotions are not relevant and targeted to each customer individually. As a result, the promotions do not respond to customers.

The Impact of Problem

Decrease in revenue, loss of customers, and inefficient promotional spending.

The Objective of Problem

Earn more revenue by increasing the number of customers who make purchases after receiving promotions.
Increase customer response to relevant and attractive product promotions.
Reduce the cost of promotions that do not deliver the desired results by directing them to more appropriate target customers.

Expected Outcomes

Increased the number of customers making purchases by 45%.
Increase customer response to product promotions by 30%.
Reduce promotion costs by 15%.

The Stakeholder

Customer, a “KlikBeli” customer is a person or individual either from an organization, group, or agency who uses the “KlikBeli” platform to purchase products. Customer satisfaction is an important factor in the success of “KlikBeli”.
Marketing team, “KlikBeli” marketing team is responsible for designing and implementing promotional marketing campaigns, developing promotional strategies, and understanding customer preferences.
Management team, the “KlikBeli” management team is responsible for strategy and decision-making related to the “KlikBeli” platform.
IT and development team, the IT and development team of “KlikBeli” conducts the development, testing, and maintenance of the system on the “KlikBeli” platform.
Partners and suppliers, partners and suppliers who cooperate with “KlikBeli” in providing products and services.
Investors, “KlikBeli” investors are parties who have an investment or financial interest in the “KlikBeli” company.

Objectives of Each Stakeholder

The customer’s objective is to get relevant promotions and special offers, obtain quality products, get easy and convenient access in using the “KlikBeli” platform.
Marketing team’s objective is to enhance “KlikBeli” strong brand image and good reputation in the market through promotion/advertising, attract new customers and encourage existing customers to make further purchases, optimize ROI (Return on Investment) of promotional campaigns.
The management team’s objective is to achieve sustainable business growth and higher profits.
The IT and development team’s objective is to improve the functionality and performance of the platform, develop and maintain a platform that is used by customers easily and add attractive features.
Partner and supplier objectives are to maintain a mutually beneficial working relationship with “KlikBeli”, provide quality products, and increase sales through the “KlikBeli” platform.
The investor’s objective is to see the growth and value of the company, get a profit from investing in the company “KlikBeli”.

Limitation of Related Issues

Some of the data limitations are :

Customer data limitations including privacy and data protection policies must be followed by the company “KlikBeli” in collecting, using, and storing customer data. Customer data includes personal information such as name, address, phone number, email address, and purchase history.
The limitation of promotional data relates to the sustainability of promotions. The available promotion budget and promotion policy must be followed by “KlikBeli” in terms of using customer data for promotional purposes. Promotional data includes information such as the type of promotion, promotion period, type of products offered, and promotional methods used.
The limitation of analytical data is related to the ability of “KlikBeli” company in collecting, processing, analyzing data and the availability of resources and technology for complex data analysis. Analytical data includes data obtained from customer analysis and purchase trends.

Then there are some inventory limitations :

Inventory limitations are related to the variety of products available. If the company has limitations in the variety of products offered, for example, limited product categories or brand variations, the promotions will be limited to the products available, so it may be difficult to target customers with different preferences.
Inventory limitations relate to the amount of product inventory available in a company’s warehouse or distribution center. If the company has a limited amount of stock, especially for the most in-demand or popular products then it can be a limitation in meeting customer demand or providing offers through promotions.
Inventory limitations are related to changes in market trends. If a company cannot predict changes in trends or anticipate seasonal demand with sufficient inventory then there can be an imbalance between promotions conducted and the availability of relevant products.

Furthermore, there are human resource limitations :

The main human resource limitation is the number of employees available in the company. If a company has a limited marketing team, it will be difficult to implement a broad and complex promotional strategy. A limited number of employees can limit the company’s ability to effectively manage and execute various promotional campaigns.

In addition, there are human resource limitations related to the skills and knowledge possessed by the marketing team. If the marketing team does not have a deep understanding of customer buying behavior and preferences, it will be difficult to design targeted and attractive promotions. Lack of specialized skills, such as data analysis or digital platform management can also limit a company’s ability to maximize the use of data and digital promotional tools.

Furthermore, human resource limitations are related to employee training and development. If a company does not have adequate training programs or lacks support for employee skill development, then the marketing team may not have the latest knowledge of effective promotional practices. Lack of training and development can limit a company’s ability to meet promotional challenges and drive innovation in marketing strategies.

Some Potential Options or Candidate Solutions

Non-ML solutions :

Conduct regular market surveys or research to understand customer buying behavior and trends. This can be done through questionnaires, interviews or data analysis from external sources.
Send special offers based on customer purchase history.
Utilize relevant social media or online advertising as promotional channels that match the characteristics of the target market.
Measure and evaluate the performance of each promotional campaign to evaluate its success. This helps to know the customer response so that the company can make the necessary improvements and adjustments.

ML solutions :

Using machine learning algorithms such as collaborative filtering, content-based filtering or hybrid filtering. The recommendation system will utilize existing customer data, product data, and transaction data to perform calculations and generate product recommendations tailored to each customer’s preferences. Recommendations can be displayed in various places within the “KlikBeli” platform, such as the home page, related product pages, or in promotional emails sent to customers.
Using machine learning techniques such as association rule mining to identify hidden buying patterns among products purchased by customers. This helps companies to offer relevant and attractive product packages or cross-product promotions.
Using machine learning methods such as time series analysis or forecasting to predict future product trends and demand. This can help companies optimize product inventory and set up effective promotions in the face of market changes.

The Business Process

In the business process of the problem, the process starts with a “customer connected ” to the KlikBeli platform. After that, there is a “customer interaction” with the platform, where they browse products and make purchases. Next, a “analyze customer preference process” is performed to understand the customer’s preferences and purchasing behavior. In this process, the system uses a hybrid recommendation method, which combines two or more recommendation methods, such as collaborative filtering and content-based filtering. This aims to produce more accurate and relevant recommendations for each customer.

After the customer preference analysis process, there are two different paths using the “XOR split gateway”. First, there is the “generate general offer process” where product offers are sent out in general to all customers. However, since these offers are not personalized, there is a possibility of customers becoming less interested and feeling that the offer is not relevant to them. Secondly, there is the “generate personalized offer process” where product offers are personalized based on customer preferences and purchase behavior. In this process, the system uses pre-analyzed information to generate offers that are more relevant and attractive to each customer. These personalized offers are then sent to relevant customers, with the hope that they will respond positively to the offer and make a purchase.

By using a hybrid recommendation method, the system can combine the advantages of different recommendation approaches and provide a more personalized and relevant experience for customers. This is expected to increase the response rate of customers to offers and promotional campaigns and encourage them to make further purchases on the KlikBeli platform.

The Solution Process

The process of finding solutions to overcome the challenges of “KlikBeli” promotion can go through the following steps.

Identify problems such as low customer response to promotions or difficulty in targeting promotions effectively.
Next, collect data such as transaction data, customer data, product reviews or other data that can provide insight into customer buying behavior.
The collected data is then processed and cleaned. This step involves handling missing values, data normalization, removal of duplicate data and data transformation if required.
Then do data exploration. This involves statistical analysis, data visualization and discovery of patterns or relationships contained in the data. The goal is to understand the characteristics of the data, identify trends and find valuable information that can be used.
After a good understanding of the data, the next step is to select appropriate machine learning techniques. The choice of techniques includes a recommendation system, which is then developed using appropriate algorithms and trained using prepared data to identify customer patterns and preferences in order to provide relevant recommendations.
Once a machine learning model is developed, the next step is to evaluate its performance. Evaluation metrics are used such as precision or recall. These evaluations help to understand the performance of the model and measure how well it can provide accurate and relevant recommendations.
The evaluated recommendation model is implemented on the “KlikBeli” system and thoroughly tested to ensure its good performance. This test involves simulating the use of the recommendation system by real users to see the extent to which the model can provide recommendations that meet expectations.
Once the recommendation system is implemented, it is important to periodically monitor its performance. If weaknesses or opportunities for improvement are found, the model is iterated and refined to improve the quality of recommendations.

Pros and Cons of Candidate Solutions

Pros :

By using machine learning techniques of recommendation system, the company “KlikBeli” can provide personalized recommendations according to each customer’s preferences.
By providing a better shopping experience through accurate recommendations, “KlikBeli” can increase customer retention and build loyalty.
The recommendation system can help “KlikBeli” optimize their marketing strategy by directing promotional efforts to customers who are most likely to make a purchase.

Cons :

Machine learning solutions require accurate and representative data to provide optimal results. If the data used is incomplete, inaccurate, or not perfectly representative of the customer population, the results of the machine learning model may be biased or ineffective.
Not all recommendations provided by the system may match customer preferences. Errors in recommendations can reduce customer trust and reduce the effectiveness of the system.
The recommendation system used can be limited by the capabilities of the algorithms used. If the algorithms are not effective enough, the quality of recommendations may be affected.
Implementation of machine learning solutions requires adequate infrastructure and technical capabilities. Processing large data, training models, and running machine learning algorithms require sufficiently powerful computing resources and appropriate systems.

The impact of the Candidate Solution

By providing relevant product recommendations to customers, the likelihood of additional purchases will increase. This will result in increased revenue and business growth for “KlikBeli”.
“KlikBeli” customers will have a more personalized and satisfying shopping experience. Customers will feel cared for and directed to products that suit their needs and interests.
“KlikBeli” can build stronger relationships with customers. Relevant product recommendations will make customers more likely to keep shopping on the “KlikBeli” platform rather than switching to competitors.
“KlikBeli” can optimize marketing strategies. By understanding customers’ preferences and purchasing behavior, “KlikBeli” can direct their promotional efforts more efficiently and allocate marketing resources better.
“KlikBeli” can make better business decisions. Information on customer preferences and behavior can help in new product development, pricing, marketing strategies, and more effective stock management.

This project must have

Personalised product recommendation features such as collaborative filtering, content-based filtering or hybrid filtering, will help “KlikBeli” display products that are most relevant to customers’ interests. This can increase purchase opportunities and strengthen customer engagement.
The project must be integrated with the existing KlikBeli platform seamlessly. It should be able to retrieve customer data, process it, and deliver recommendations within the platform.
The project should have a user-friendly interface that displays personalized recommendations to the customers. It should be visually appealing and easy to navigate.

This project nice to have

Implementing real-time recommendation capabilities would allow customers to receive personalized recommendations instantly as they browse the platform, enhancing the overall user experience.
Integrating the recommendation system across multiple channels, such as mobile apps, social media platforms, and email marketing, would enable consistent and personalized recommendations across various touchpoints.
Incorporating the ability to offer seasonal promotions and trend-based recommendations would allow KlikBeli to capitalize on changing consumer preferences and boost sales during specific periods.

Not covered by this project

The project does not directly cover the payment and checkout process, including transaction security, or streamlining the checkout experience.
The project does not specifically address fraud detection or ensuring the integrity of customer data.
The project does not specifically focus on improving customer service or addressing customer inquiries, complaints, or support requests. Customer service is a separate area that requires dedicated resources and processes.

Timeline Project

Milestones and Deadline

Potential Risk

If the customer data used in the development is not representative enough or does not include relevant information, the analysis and prediction results may be inaccurate. This may reduce the effectiveness of promotional personalization.
Machine learning modeling can produce errors and biases that affect the quality of predictions. These errors can lead to offers that are irrelevant or less appealing to customers, thus harming the effectiveness of promotional campaigns.
In collecting and using customer data, there is a risk of data privacy and security breaches. If customer data is not properly processed and stored, or data leaks occur that reveal customers’ personal information, this can damage customer trust and harm the company’s reputation.

Tools Needed

To analyze and visualize customer data, data visualization tools such as Tableau, Power BI, or Matplotlib (for Python) can be used. This will help in understanding customer buying patterns and identifying relevant trends.
In managing these projects, team collaboration tools such as Slack, Trello, or Jira can be used to communicate, organize tasks, and track project progress.
In implementing the promotion personalization feature, machine learning tools are needed that can be used to build and train predictive models. Some tools in the field of machine learning include Python with libraries such as scikit-learn, TensorFlow, or PyTorch.
To manage large and complex customer data, efficient database systems and data warehousing are required. Examples of commonly used databases include MySQL, PostgreSQL, or MongoDB. Data warehousing can be implemented using tools such as Apache Hadoop or Apache Spark.
Given the volume and complexity of the data at hand, the use of cloud computing services can help with data storage, fast processing, and scalability. Some cloud computing providers are Amazon Web Services (AWS), Microsoft Azure, or Google Cloud Platform.

Reference Materials

Hansson L. 2015. Product Recommendations in E-commerce Systems using Content-based Clustering and Collaborative Filtering. Master’s thesis. Apptus Technologies AB.

Liao M., Sundar S. S. 2022. When E-Commerce Personalization Systems Show and Tell: Investigating the Relative Persuasive Appeal of Content-Based versus Collaborative Filtering. Journal of Advertising. 2022, VOL. 51, NO. 2, 256–26.

Ekstrand M. D. 2019. Recommender Systems. Lecture Notes in Boise State University.

CSE 255-Lecture 5. Data Mining and Predictive Analytics Recommender Sytems. Lecture Notes in University of California San Diego.

Husein A. T. A., Rahma A. M. S., Wahab H. B. A. 2021. Recommendation System for E-Commerce System An Overview. Journal of Physics: Conference Series.

Sivapalan S. Sadeghian A. Rahanam H. 2014. Recommender Systems in E-commerce. Ryeson University.

Bahamonde A. 2013. Recommendation System. Lecture Notes in Cornell University.